GPT-6 LunavsGPT-6 Sol
Across 4 shared benchmarks, GPT-6 Sol leads overall: GPT-6 Luna wins 0, GPT-6 Sol wins 4, with 0 ties and an average score difference of -6.58.
GPT-6 Luna
OpenAI · 2026-09-22 · Reasoning model
GPT-6 Sol
OpenAI · 2026-09-22 · Reasoning model
GPT-6 Luna0 wins(0%)(100%)4 winsGPT-6 Sol
Benchmark scores
Grouped by capability, sorted by largest gap within each. 4 shared benchmarks.
Coding and Software Engineer
GPT-6 Sol 2/2| Benchmark | GPT-6 Luna | GPT-6 Sol | Diff |
|---|---|---|---|
| FrontierCode 1.1 Main | 42.409 / 9Max (With Tools) | 49.307 / 9Max (With Tools) | -6.90 |
| DeepSWE | 66.6037 / 91Max (With Tools) | 68.8025 / 91Max (With Tools) | -2.20 |
Agent Level Benchmark
GPT-6 Sol 1/1| Benchmark | GPT-6 Luna | GPT-6 Sol | Diff |
|---|---|---|---|
| Agents' Last Exam | 50.905 / 24Max (With Tools) | 56.402 / 24Max (With Tools) | -5.50 |
AI Agent - Tool Usage
GPT-6 Sol 1/1| Benchmark | GPT-6 Luna | GPT-6 Sol | Diff |
|---|---|---|---|
| OSWorld 2.0 | 52.7011 / 13Max (With Tools) | 64.407 / 13Max (With Tools) | -11.70 |
Specs
| Field | GPT-6 Luna | GPT-6 Sol |
|---|---|---|
| Publisher | OpenAI | OpenAI |
| Release date | 2026-09-22 | 2026-09-22 |
| Model type | Reasoning model | Reasoning model |
| Architecture | Dense | Dense |
| Parameters | Not available | Not available |
| Context length | 1.05M | 1.05M |
| Max output | 128K | 128K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GPT-6 Luna | GPT-6 Sol |
|---|---|---|
| Text input | $0.1 / 1M tokens | $2 / 1M tokens |
| Text output | $0.5 / 1M tokens | $10 / 1M tokens |
| Cache read | $0.01 / 1M tokens | $0.2 / 1M tokens |
| Cache write | $0.125 / 1M tokens | $2.5 / 1M tokens |
Summary
- GPT-6 Solleads in:Coding and Software Engineer (2/2), Agent Level Benchmark (1/1), AI Agent - Tool Usage (1/1)
On average across the 4 shared benchmarks, GPT-6 Sol scores 6.58 higher.
Largest single-benchmark gap: OSWorld 2.0 — GPT-6 Luna 52.70 vs GPT-6 Sol 64.40 (-11.70).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.